Copilot ROI: A 60-User Model You Can Recalculate

Every Copilot business case I have been handed had the same defect: a savings number with no arithmetic behind it. Somebody assumed two hours a week per user, multiplied by a headcount, and the slide was done. Nobody could say where the two hours came from, how many of those users would still be opening Copilot in month four, or what happens to the number when half the company's SharePoint estate is excluded from search.

So here is the model instead of the slide. Sixty users, one year, every input named and every subtotal shown, using the two productivity constants Microsoft itself publishes. Substitute your own numbers where mine are guesses. That is the point.

Correction: this replaces an earlier version of this article. That version carried a $195,000 figure that did not follow from its own tables, and a 1.4:1 / 5.4:1 headline that was not derived from anything. Both are gone. What follows is recalculated from scratch.

The two numbers Microsoft publishes

You do not have to invent productivity assumptions. Microsoft ships two of them inside its own reporting, and it documents where they come from.

$72 per hour. The Copilot Dashboard in Viva Insights converts Copilot assisted hours into Copilot assisted value by multiplying by an average hourly rate. Microsoft's default is $72, based on US Bureau of Labor Statistics data on employer costs for employee compensation. It is a fully loaded employer cost, not a salary, and it is editable in the dashboard's value calculator.

Six minutes per assisted task. The Copilot Studio agents report uses a default time savings multiplier of six minutes per knowledge reference, drawn from research by Microsoft's Office of the Chief Economist on information-retrieval and writing tasks. Same six minutes for both categories.

Microsoft is careful about what these are. Learn calls the assisted-hours metric "a general estimate" and "broad approximations based on the best available research rather than precise calculations," and says the methodology will change as new research lands. Treat them the same way: as a defensible starting point that someone else has already had to justify, not as measurement.

Put them together with a working year and you get one clean constant. At 220 working days (my assumption, adjust for your leave policy), one assisted action per user per working day is:

220 days × 6 minutes ÷ 60 = 22 hours per user per year, worth 22 × $72 = $1,584.

That single line turns "will Copilot pay off" into a question you can answer from a usage report: how many assisted actions per day, by how many people.

What the licence actually costs

Microsoft Learn does not publish prices, so this comes from the public pricing pages, checked on 6 September 2026, in USD, commercial, United States. Prices differ by region and by agreement, so check yours before you present anything.

• Microsoft 365 Copilot (the enterprise add-on): $30.00 user/month, paid yearly, annual commitment.
• Microsoft 365 Copilot Business (the add-on for the Business plans): listed at $21.00 user/month paid yearly, shown at a promotional $18.00 on the day I checked.
• Microsoft 365 Copilot Chat: included at no additional cost for every Entra account with an eligible Microsoft 365 or Office 365 subscription.

I model the $30 figure because the tenant in this example runs on Microsoft 365 E3. A Business Premium tenant pays less, which improves every ratio below.

Two licensing facts that break more pilots than the price does. First, the eligible base plans are far broader than E3 and E5: Business Basic, Standard and Premium, E3/E5/E7, F1/F3, A3/A5, Office 365 E1/E3/E5/F3 and the Teams plans all qualify. Business Premium is the realistic entry point for an SME. Second, and this is a hard blocker: the user's primary mailbox must be in Exchange Online. On-premises and hybrid mailboxes do not support mailbox grounding, so a licensed user with an on-premises mailbox gets Copilot for cloud content and nothing from their mail and calendar. Find those users before you count them in the business case.

The cost side, line by line

The scenario: a 500-person manufacturer on Microsoft 365 E3 rolls Copilot out to 60 people across management, sales and engineering. Rates marked as assumptions are mine and should be replaced with yours.

Cost item	How it is calculated	Year 1
Copilot licences	60 users × $30/month × 12	$21,600
Permission and findability remediation (external)	100 h × $160/h (assumed rate)	$16,000
Training, two hours per user (internal time)	60 users × 2 h × $72/h	$8,640
Admin time for adoption reporting	4 h/month × 12 × $72/h	$3,456
Total, year 1		$49,696
Recurring, year 2 onward	Licences + admin time	$25,056

What is deliberately not in the table. Copilot connectors cost nothing extra to index: Learn states that indexing of synced connector data incurs no extra cost for tenants with Microsoft 365 licences, and connector grounding is included with a Copilot licence. SharePoint Advanced Management is included with the Copilot licence too, which is why the remediation line is labour and not software. Purview DLP for the Copilot location is the one place a real licence gap can appear: blocking Copilot from processing labelled files and emails needs Microsoft 365 E5 or the Purview Suite, while DLP on prompts is available to all Copilot and Copilot Chat users. If you need the file-level control and sit on E3, add that line yourself.

The benefit side, line by line

One formula, four inputs, nothing hidden:

Assisted hours = active users × actions per working day × 220 days × 6 minutes ÷ 60
Value = assisted hours × $72

Two of those inputs are Microsoft's published defaults and one is a calendar assumption. The remaining two are the only things you genuinely have to estimate, and both can later be read out of the Copilot usage report instead of argued about: how many licensed users are actually active, and how many assisted actions each takes per working day. If a Copilot business case cannot be reduced to those two unknowns, the number in it is decoration.

Three scenarios and a break-even

Same cost base of $49,696 in all three. Only the two unknowns change.

Scenario	Active users	Actions/day	Assisted hours	Value at $72/h	Net, year 1	Ratio
Licences handed out, nothing else done	24 (40%)	1	528	$38,016	−$11,680	0.77:1
Findability fixed, users trained	45 (75%)	2	1,980	$142,560	+$92,864	2.87:1
Mature, content maintained	54 (90%)	4	4,752	$342,144	+$292,448	6.88:1

The first row is a loss, and it is the most common outcome I see in the first year. Forty per cent of licensed users touching Copilot roughly once a working day is not a strawman; it is what a rollout with no enablement and a messy SharePoint estate produces.

Break-even for the middle scenario: one-off costs of $24,640 (remediation plus training), recurring costs of $2,088 a month, benefit of $11,880 a month, so $9,792 net a month against $24,640 of setup. That clears at 2.5 months, on the flattering assumption that the benefit runs at full rate from day one. It never does. Assume a three-month ramp and you land closer to four months.

Year two is where the shape changes. The remediation and training are spent, so the middle scenario becomes $142,560 against $25,056 of recurring cost, or 5.69:1. That only holds if someone keeps owning the content. The structure decays; the licence bill does not.

Where this model is weakest

Three things, and I would rather say them than have a CFO find them.

Assisted hours are not cash. An hour that Copilot gives back becomes value only if it is spent on something the business would otherwise have paid for. Microsoft's own framing is that employees "re-invest" the time. In a sales team with a pipeline gap, that reinvestment is real. In a back-office function with no queue, it is slack. If you think only half the freed time converts, halve every benefit figure: the middle scenario drops from 2.87:1 to 1.43:1 and break-even moves from 2.5 to 6.4 months. Still positive, much less exciting, and much easier to defend.

Six minutes is an average of research results, not a measurement of your tenant. Microsoft says so plainly and reserves the right to change it. If your users mostly draft short emails, it is generous. If they spend their day hunting through document libraries, it is probably conservative.

The two unknowns are estimates until you have real usage data. Which is the argument for running Copilot Chat first, seeing what demand looks like in the free tier, and buying licences against evidence rather than a forecast.

Why findability moves the number more than the licence

The reason knowledge management belongs in a financial article is mechanical, not philosophical. Copilot's value is the number of grounded, useful answers it produces. That number is capped by what Copilot is allowed to retrieve for the person asking. Three documented controls set that cap, and two of them are things your own admins switch on.

Permissions are the boundary, and there is no way around them. The semantic index is built for the whole tenant, but it only surfaces a result to a user who already has access to that content through role-based access control. Copilot prompts run in the security context of the person asking. This cuts both ways: an over-shared site is an exposure problem, and an under-shared one is an ROI problem. I wrote up the audit side of that in the SharePoint permission audit for Copilot readiness.

A site excluded from search is invisible to Copilot. Setting a site's "Search and offline availability" to no removes it from Microsoft Search and from the semantic index at the same time. There is no option to exclude one and keep the other. Somebody who ticked that box on the engineering wiki three years ago has quietly deleted a chunk of your business case.

Restricted Content Discovery is a governance tool with a direct price. It is a site-level setting, included with the Copilot licence through SharePoint Advanced Management, that keeps a site out of organisation-wide search and Copilot responses while you review its permissions. It changes no permissions and removes nothing from the search index, but it does strip the AI entry points out of the site: no Copilot button, no AI actions menu, no create-pages-with-AI. Microsoft's own caution is unusually direct: excessive use "can reduce the amount of content available to organization-wide search and Microsoft Copilot experiences, which can affect the completeness and relevance of search results and AI-generated responses." It applies to SharePoint sites only, not OneDrive, and on sites with more than 500,000 items the change can take over a week to propagate.

# Restrict one site from org-wide search and Copilot discovery
Set-SPOSite -Identity https://contoso.sharepoint.com/sites/finance -RestrictContentOrgWideSearch $true

# Let site admins manage the setting for their own sites (they must supply a justification)
Set-SPOTenant -DelegateRestrictedContentDiscoverabilityManagement $true

If you are still using Restricted SharePoint Search, the allow-list approach, plan your way off it. It is retiring, new enablement is blocked from 31 July 2026, and Microsoft directs customers to Restricted Content Discovery instead.

There is an upside lever too, and it is cheap. When a user attaches a SharePoint document library or folder to a prompt, or pastes its URL, Copilot uses that library's column metadata as an additional signal to constrain and rank results. Consistent metadata on the libraries people actually work in is one of the few knowledge-management chores with a visible effect on answer quality.

The remediation tooling is already paid for. A Copilot licence includes SharePoint Advanced Management, which is where the data access governance reports, the "everyone except external users" insights, site permission reports, site access review, the SharePoint Admin Agent and the content management assessment live. That is why my cost table has 100 hours of labour and no software line: you are buying somebody's attention, not a product.

Measuring it in your own tenant

Replace my two unknowns with real numbers as soon as you have them.

• Copilot readiness report (Microsoft 365 admin center, Reports > Usage > Microsoft Copilot) shows which users meet the licensing prerequisites and flags technical blockers. Run it before you order anything.
• Copilot usage report, same place, gives you active users and usage trends per app, and will tell you how many licensed users were inactive over the past 28 days. Data for a given day appears within about 48 hours. This is where your "active users" input comes from, and where you find the licences to reassign.
• Copilot Chat usage report covers users without a Copilot licence. Use it as a demand signal before you buy.
• Copilot Dashboard in Viva Insights gives you assisted hours and assisted value directly, with the $72 rate editable in the calculator. Data appears within about seven days of licences being assigned.

Two gotchas worth knowing before you promise a board a number. Copilot assisted hours needs optional diagnostic data enabled to be complete; with only required diagnostic data the figure is understated. And the Copilot business impact report, the one that correlates usage against your own outcome metrics, wants at least 65 Copilot users and 65 non-Copilot users to run properly. A 60-licence rollout is below that line. Plan the pilot cohort accordingly if the correlation analysis matters to you.

What I would do before buying 60 licences

1. Run the readiness report and count the users whose primary mailbox is not in Exchange Online. Remove them from the business case or fix the mailbox.
2. Turn on Copilot Chat properly, leave it for six weeks, and read the Copilot Chat usage report. Demand you can see beats demand you assumed.
3. Pull the data access governance reports and the "everyone except external users" insights. Fix the worst over-sharing; use Restricted Content Discovery only where a site genuinely needs to sit out the review, and take it off again afterwards.
4. Find every site with search disabled and decide, deliberately, whether it should stay invisible.
5. Decide whether you need DLP for the Copilot location on files and emails. If yes and you are on E3, price the Purview Suite before, not after.
6. Pick a pilot cohort with a real, boring, repetitive information-retrieval problem. That is where the six minutes actually show up.
7. Agree the rate. If $72 is wrong for your workforce, change it once, at the start, and use the same rate on both sides of the model.

Then run the numbers again with your inputs. If the result embarrasses the project, that is the model working.

Glossary

Grounding

Combining retrieved tenant content with the user's prompt before the language model answers. Without it, the model has only its training data.

Semantic index

A vectorised index built from text-based SharePoint content at tenant level, used to find conceptually related material rather than exact keyword matches. It respects existing permissions and is generated for users with a paid Copilot licence.

Copilot assisted hours

Microsoft's estimate of time returned to employees, computed from Copilot actions multiplied by research-derived factors. A general estimate, not a measurement.

Copilot assisted value

Assisted hours multiplied by an hourly rate, defaulting to $72 from US Bureau of Labor Statistics employer-cost data, editable in the dashboard calculator.

Restricted Content Discovery (RCD)

A SharePoint site setting that keeps a site out of organisation-wide search and Copilot responses without changing permissions or removing content from the index. Part of SharePoint Advanced Management.

Copilot connectors

Formerly Graph connectors. Synced connectors index external content into Microsoft Graph; federated connectors query the source at runtime without indexing and require a Copilot add-on licence per querying user.

EXTRACT usage right

The rights-management permission that allows copying text from encrypted content. Copilot needs it, together with VIEW, before it will return content from an encrypted file.

Frequently Asked Questions

How much does Microsoft 365 Copilot cost per user?

On the public pricing pages checked on 6 September 2026, in USD for the United States: Microsoft 365 Copilot, the enterprise add-on, is $30.00 per user per month paid yearly on an annual commitment. Microsoft 365 Copilot Business, the add-on for the Business plans, is listed at $21.00 per user per month paid yearly and was showing a promotional $18.00 that day. Microsoft 365 Copilot Chat is included at no extra cost with an eligible Microsoft 365 or Office 365 subscription. Microsoft Learn does not publish prices, so always confirm against your own agreement and region.

What ROI should I expect from Microsoft 365 Copilot?

There is no single figure, and any vendor quoting one has skipped the arithmetic. Using Microsoft's own published defaults of $72 per hour and six minutes saved per assisted task, one Copilot action per user per working day is worth about $1,584 a year against a $360 annual licence. In a 60-user model with $49,696 of first-year cost, year one lands anywhere from 0.77:1, a loss, when 40 per cent of users are active at one action a day, to 6.88:1 when 90 per cent are active at four. The spread is driven by adoption and by how much content Copilot is allowed to find, not by the licence price.

Do I need Microsoft 365 E5 to use Copilot?

No. The eligible base plans include Microsoft 365 Business Basic, Standard and Premium, E3, E5 and E7, F1 and F3, A3 and A5, Office 365 E1, E3, E5 and F3, and the Teams plans. Business Premium is the realistic entry point for a smaller organisation. The hard requirement is not the plan tier but the mailbox: the user's primary mailbox must be in Exchange Online, because on-premises and hybrid mailboxes do not support mailbox grounding. E5 or the Purview Suite becomes relevant only if you want Purview DLP to stop Copilot processing labelled files and emails; DLP on prompts is available to all Copilot users.

How do I measure whether Copilot is paying for itself?

Start with the Copilot usage report in the Microsoft 365 admin center for active users and inactivity over the past 28 days, with roughly 48 hours of latency. Add the Copilot Dashboard in Viva Insights for assisted hours and assisted value, and change the default $72 rate to your own. Two caveats: assisted hours are incomplete unless optional diagnostic data is enabled, and the Copilot business impact report, which correlates usage against your own outcome metrics, needs at least 65 Copilot and 65 non-Copilot users to run.

---

References

1. Microsoft Learn: Minimum requirements to deploy Microsoft Copilot — eligible base plans, Exchange Online mailbox requirement
2. Microsoft Learn: License options for Microsoft Copilot
3. Microsoft: Copilot plans and pricing — list prices, checked 6 September 2026
4. Microsoft Learn: Copilot Dashboard in Viva Insights — assisted hours and value, the $72 default rate, diagnostic data
5. Microsoft Learn: Copilot Studio agents report — the six-minute time savings multiplier and its source
6. Microsoft Learn: Copilot business impact report — 65 Copilot and 65 non-Copilot user minimum
7. Microsoft Learn: Microsoft Copilot usage report
8. Microsoft Learn: Microsoft Copilot readiness report
9. Microsoft Learn: Semantic indexing for Microsoft Copilot — permissions, excluding sites from search, column metadata
10. Microsoft Learn: Restrict discovery of SharePoint sites and content
11. Microsoft Learn: Restricted SharePoint Search — retirement and 31 July 2026 date
12. Microsoft Learn: SharePoint Advanced Management features in Microsoft Copilot licenses
13. Microsoft Learn: Prerequisites for deploying Copilot connectors — synced vs federated, indexing cost
14. Microsoft Learn: DLP for Microsoft 365 Copilot and Copilot Chat
15. Microsoft Learn: Microsoft Purview service description — which plans include DLP for Copilot files and emails
16. Microsoft Learn: How data is protected and audited in Microsoft 365 and Microsoft Copilot — sensitivity labels, encryption, EXTRACT and VIEW
17. Microsoft Learn: Get ready for Microsoft Copilot and agents with SharePoint Advanced Management